Hugging Face's logo

STMicroelectronics
/
mobilenetv2

Image Classification
Model card Files
xet
 Community
FBAGSTM commited on
Commit
9463a64
·
verified ·
1 Parent(s): ba2348d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +262 -6
README.md CHANGED
@@ -1,6 +1,262 @@
1
- ---
2
- license: other
3
- license_name: sla0044
4
- license_link: >-
5
- https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/LICENSE.md
6
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ license_name: sla0044
4
+ license_link: >-
5
+ https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/LICENSE.md
6
+ pipeline_tag: image-classification
7
+ ---
8
+ # MobileNet v2
9
+
10
+ ## **Use case** : `Image classification`
11
+
12
+ # Model description
13
+
14
+
15
+ MobileNet v2 is very similar to the original MobileNet, except that it uses inverted residual blocks with bottlenecking features.
16
+
17
+ It has a drastically lower parameter count than the original MobileNet.
18
+
19
+ MobileNet models support any input size greater than 32 x 32, with larger image sizes offering better performance.
20
+ Alpha parameter: float, larger than zero, controls the width of the network. This is known as the width multiplier in the MobileNetV2 paper, but the name is kept for consistency with applications.
21
+
22
+ If alpha < 1.0, proportionally decreases the number of filters in each layer.
23
+
24
+ If alpha > 1.0, proportionally increases the number of filters in each layer.
25
+
26
+ If alpha = 1.0, default number of filters from the paper are used at each layer.
27
+
28
+ (source: https://keras.io/api/applications/mobilenet/)
29
+
30
+ The model is quantized in int8 using tensorflow lite converter.
31
+
32
+ ## Network information
33
+
34
+
35
+ | Network Information | Value |
36
+ |-------------------------|-----------------|
37
+ | Framework | TensorFlow Lite |
38
+ | MParams alpha=0.35 | 1.66 M |
39
+ | Quantization | int8 |
40
+ | Provenance | https://www.tensorflow.org/api_docs/python/tf/keras/applications/mobilenet_v2 |
41
+ | Paper | https://arxiv.org/pdf/1801.04381.pdf |
42
+
43
+ The models are quantized using tensorflow lite converter.
44
+
45
+
46
+ ## Network inputs / outputs
47
+
48
+
49
+ For an image resolution of NxM and P classes
50
+
51
+ | Input Shape | Description |
52
+ | ----- | ----------- |
53
+ | (1, N, M, 3) | Single NxM RGB image with UINT8 values between 0 and 255 |
54
+
55
+ | Output Shape | Description |
56
+ | ----- | ----------- |
57
+ | (1, P) | Per-class confidence for P classes in FLOAT32|
58
+
59
+
60
+ ## Recommended platforms
61
+
62
+
63
+ | Platform | Supported | Recommended |
64
+ |----------|-----------|-----------|
65
+ | STM32L0 |[]|[]|
66
+ | STM32L4 |[x]|[]|
67
+ | STM32U5 |[x]|[]|
68
+ | STM32H7 |[x]|[x]|
69
+ | STM32MP1 |[x]|[x]|
70
+ | STM32MP2 |[x]|[x]|
71
+ | STM32N6 |[x]|[x]|
72
+
73
+ # Performances
74
+
75
+ ## Metrics
76
+
77
+ Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
78
+
79
+
80
+ ### Reference **NPU** memory footprint on food-101 and ImageNet dataset (see Accuracy for details on dataset)
81
+ |Model | Dataset | Format | Resolution | Series | Internal RAM | External RAM | Weights Flash | STM32Cube.AI version | STEdgeAI Core version |
82
+ |----------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
83
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | food-101 | Int8 | 128x128x3 | STM32N6 | 240 | 0.0 | 715.67 | 10.0.0 | 2.0.0 |
84
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6 | 980 | 0.0 | 730.7 | 10.0.0 | 2.0.0 |
85
+ | [MobileNet v2 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_1.0_224_fft/mobilenet_v2_1.0_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6 | 2058 | 0.0 | 3110.05 | 10.0.0 | 2.0.0 |
86
+ | [MobileNet v2 0.35 fft](ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Person | Int8 | 128x128x3 | STM32N6 | 240 | 0.0 | 589.45 | 10.0.0 | 2.0.0 |
87
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_128/mobilenet_v2_0.35_128_int8.tflite) | ImageNet | Int8 | 128x128x3 | STM32N6 | 240 | 0.0 | 1840.94 | 10.0.0 | 2.0.0 |
88
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_224/mobilenet_v2_0.35_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6 | 980 | 0.0 | 1855.97 | 10.0.0 | 2.0.0 |
89
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6 | 2058 | 0.0 | 4235.31 | 10.0.0 | 2.0.0 |
90
+ | [MobileNet v2 1.4](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.4_224/mobilenet_v2_1.4_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6 | 2361 | 0.0 | 7315.69 | 10.0.0 | 2.0.0 |
91
+
92
+
93
+ ### Reference **NPU** inference time on food-101 and ImageNet dataset (see Accuracy for details on dataset)
94
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
95
+ |--------|------------------|--------|-------------|------------------|------------------|---------------------|-------|----------------------|-------------------------|
96
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | food-101 | Int8 | 128x128x3 | STM32N6570-DK | NPU/MCU | 3.33 | 300.30 | 10.0.0 | 2.0.0 |
97
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 6.12 | 163.40 | 10.0.0 | 2.0.0 |
98
+ | [MobileNet v2 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_1.0_224_fft/mobilenet_v2_1.0_224_fft_int8.tflite) | food-101 | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 18.08 | 55.32 | 10.0.0 | 2.0.0 |
99
+ | [MobileNet v2 0.35 fft](ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Person | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 2.99 | 334.45 | 10.0.0 | 2.0.0 |
100
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_128/mobilenet_v2_0.35_128_int8.tflite) | ImageNet | Int8 | 128x128x3 | STM32N6570-DK | NPU/MCU | 6.35 | 157.48 | 10.0.0 | 2.0.0 |
101
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_224/mobilenet_v2_0.35_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 9.14 | 109.40 | 10.0.0 | 2.0.0 |
102
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 21.08 | 47.44 | 10.0.0 | 2.0.0 |
103
+ | [MobileNet v2 1.4](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.4_224/mobilenet_v2_1.4_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32N6570-DK | NPU/MCU | 35.34 | 28.30 | 10.0.0 | 2.0.0 |
104
+
105
+
106
+ ### Reference **MCU** memory footprint based on Flowers and ImageNet dataset (see Accuracy for details on dataset)
107
+
108
+ | Model | Dataset | Format | Resolution | Series | Activation RAM | Runtime RAM | Weights Flash | Code Flash | Total RAM | Total Flash | STM32Cube.AI version |
109
+ |--------------------------------------------------------------------------------------------------------------------------------------------------|----------|--------|-------------|---------|----------------|-------------|---------------|------------|------------|-------------|----------------------|
110
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Flowers | Int8 | 128x128x3 | STM32H7 | 237.32 KiB | 30.14 KiB | 406.86 KiB | 108.29 KiB | 267.46 KiB | 515.15 KiB | 10.0.0 |
111
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Flowers | Int8 | 224x224x3 | STM32H7 | 832.64 KiB | 30.19 KiB | 406.86 KiB | 108.40 KiB | 862.83 KiB | 515.26 KiB | 10.0.0 |
112
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_128/mobilenet_v2_0.35_128_int8.tflite) | ImageNet | Int8 | 128x128x3 | STM32H7 | 237.32 KiB | 30.14 KiB | 1654.5 KiB KiB | 108.29 KiB | 267.46 KiB | 1762.79 KiB | 10.0.0 |
113
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_224/mobilenet_v2_0.35_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32H7 | 832.64 KiB | 30.19 KiB | 1654.5 KiB | 108.40 KiB | 862.83 KiB | 1762.9 KiB | 10.0.0 |
114
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32H7 | 1727.02 KiB | 30.19 KiB | 3458.97 KiB | 157.37 KiB | 1757.21 KiB | 3616.34 KiB | 10.0.0 |
115
+ | [MobileNet v2 1.4](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.4_224/mobilenet_v2_1.4_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32H7 | 2332.2 KiB | 30.19 KiB | 6015.34 KiB | 191.16 KiB | 2362.39 KiB | 6206.53 KiB | 10.0.0 |
116
+
117
+ ### Reference **MCU** inference time based on Flowers and ImageNet dataset (see Accuracy for details on dataset)
118
+
119
+ | Model | Dataset | Format | Resolution | Board | Execution Engine | Frequency | Inference time (ms) | STM32Cube.AI version |
120
+ |--------------------------------------------------------------------------------------------------------------------------------------------------|----------|--------|-------------|------------------|------------------|-------------|---------------------|-----------------------|
121
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Flowers | Int8 | 128x128x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 94.34 ms | 10.0.0 |
122
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Flowers | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 290.75 ms | 10.0.0 |
123
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_128/mobilenet_v2_0.35_128_int8.tflite) | ImageNet | Int8 | 128x128x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 116.13 ms | 10.0.0 |
124
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_224/mobilenet_v2_0.35_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 313.92 ms | 10.0.0 |
125
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 1106.64 ms | 10.0.0 |
126
+ | [MobileNet v2 1.4](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.4_224/mobilenet_v2_1.4_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | STM32H747I-DISCO | 1 CPU | 400 MHz | 2010.66 ms | 10.0.0 |
127
+
128
+ ### Reference **MPU** inference time based on Flowers and ImageNet dataset (see Accuracy for details on dataset)
129
+ | Model | Dataset | Format | Resolution | Quantization | Board | Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version | Framework |
130
+ |--------------------------------------------------------------------------------------------------------------------------------------------------|----------|--------|------------|---------------|-------------------|------------------|-----------|---------------------|-------|-------|------|--------------------|-----------------------|
131
+ | [MobileNet v2 1.0_per_tensor](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8_per_tensor.tflite) | ImageNet | Int8 | 224x224x3 | per-tensor | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 11.92 ms | 92.74 | 7.26 |0 | v5.1.0 | OpenVX |
132
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | ImageNet | Int8 | 224x224x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 76.29 ms | 3.13 | 96.87 |0 | v5.1.0 | OpenVX |
133
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Flowers | Int8 | 224x224x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 25.51 ms | 4.37 | 95.63 |0 | v5.1.0 | OpenVX |
134
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Flowers | Int8 | 128x128x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 800 MHz | 9.14 ms | 12.06 | 87.94 |0 | v5.1.0 | OpenVX |
135
+ | [MobileNet v2 1.0_per_tensor](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8_per_tensor.tflite) | ImageNet | Int8 | 224x224x3 | per-tensor | STM32MP157F-DK2 | 2 CPU | 800 MHz | 332.9 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
136
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | ImageNet | Int8 | 128x128x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 194.1 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
137
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Flowers | Int8 | 224x224x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 54.52 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
138
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Flowers | Int8 | 128x128x3 | per-channel | STM32MP157F-DK2 | 2 CPU | 800 MHz | 17.16 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
139
+ | [MobileNet v2 1.0_per_tensor](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8_per_tensor.tflite) | ImageNet | Int8 | 224x224x3 | per-tensor | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 415.7 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
140
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | ImageNet | Int8 | 128x128x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 308.80 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
141
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Flowers | Int8 | 224x224x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 54.85 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
142
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Flowers | Int8 | 128x128x3 | per-channel | STM32MP135F-DK2 | 1 CPU | 1000 MHz | 27.17 ms | NA | NA |100 | v5.1.0 | TensorFlowLite 2.11.0 |
143
+
144
+ ** **To get the most out of MP25 NPU hardware acceleration, please use per-tensor quantization**
145
+ ### Accuracy with Flowers dataset
146
+
147
+
148
+ Dataset details: [link](http://download.tensorflow.org/example_images/flower_photos.tgz) , License [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/) , Quotation[[1]](#1) , Number of classes: 5, Number of images: 3 670
149
+
150
+ | Model | Format | Resolution | Top 1 Accuracy |
151
+ |-------|--------|------------|----------------|
152
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs.h5) | Float | 128x128x3 | 87.06 % |
153
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs_int8.tflite) | Int8 | 128x128x3 | 87.47 % |
154
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl.h5) | Float | 128x128x3 | 88.15 % |
155
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl_int8.tflite) | Int8 | 128x128x3 | 88.01 % |
156
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft.h5) | Float | 128x128x3 | 91.83 % |
157
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Int8 | 128x128x3 | 91.01 % |
158
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_tfs/mobilenet_v2_0.35_224_tfs.h5) | Float | 224x224x3 | 88.69 % |
159
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_tfs/mobilenet_v2_0.35_224_tfs_int8.tflite) | Int8 | 224x224x3 | 88.83 % |
160
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_tl/mobilenet_v2_0.35_224_tl.h5) | Float | 224x224x3 | 88.96 % |
161
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_tl/mobilenet_v2_0.35_224_tl_int8.tflite) | Int8 | 224x224x3 | 88.01 % |
162
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft.h5) | Float | 224x224x3 | 93.6 % |
163
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/flowers/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Int8 | 224x224x3 | 92.78 % |
164
+
165
+
166
+ ### Accuracy with Plant-village dataset
167
+
168
+
169
+ Dataset details: [link](https://data.mendeley.com/datasets/tywbtsjrjv/1) , License [CC0 1.0](https://creativecommons.org/publicdomain/zero/1.0/), Quotation[[2]](#2) , Number of classes: 39, Number of images: 61 486
170
+
171
+ | Model | Format | Resolution | Top 1 Accuracy |
172
+ |-------|--------|------------|----------------|
173
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs.h5) | Float | 128x128x3 | 99.86 % |
174
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs_int8.tflite) | Int8 | 128x128x3 | 99.83 % |
175
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl.h5) | Float | 128x128x3 | 93.51 % |
176
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl_int8.tflite) | Int8 | 128x128x3 | 92.33 % |
177
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft.h5) | Float | 128x128x3 | 99.77 % |
178
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Int8 | 128x128x3 | 99.48 % |
179
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_224_tfs/mobilenet_v2_0.35_224_tfs.h5) | Float | 224x224x3 | 99.86 % |
180
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_224_tfs/mobilenet_v2_0.35_224_tfs_int8.tflite) | Int8 | 224x224x3 | 99.81 % |
181
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_224_tl/mobilenet_v2_0.35_224_tl.h5) | Float | 224x224x3 | 93.62 % |
182
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_224_tl/mobilenet_v2_0.35_224_tl_int8.tflite) | Int8 | 224x224x3 | 92.8 % |
183
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft.h5) | Float | 224x224x3 | 99.95 % |
184
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/plant-village/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Int8 | 224x224x3 | 99.68 % |
185
+
186
+
187
+ ### Accuracy with Food-101 dataset
188
+
189
+ Dataset details: [link](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/) , License [-](), Quotation[[3]](#3) , Number of classes: 101 , Number of images: 101 000
190
+
191
+ | Model | Format | Resolution | Top 1 Accuracy |
192
+ |-------|--------|------------|----------------|
193
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs.h5) | Float | 128x128x3 | 64.22 % |
194
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs_int8.tflite) | Int8 | 128x128x3 | 63.41 % |
195
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl.h5) | Float | 128x128x3 | 44.74 % |
196
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl_int8.tflite) | Int8 | 128x128x3 | 42.01 % |
197
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft.h5) | Float | 128x128x3 | 64.22 % |
198
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Int8 | 128x128x3 | 63.41 % |
199
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_tfs/mobilenet_v2_0.35_224_tfs.h5) | Float | 224x224x3 | 72.31 % |
200
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_tfs/mobilenet_v2_0.35_224_tfs_int8.tflite) | Int8 | 224x224x3 | 72.05 % |
201
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_tl/mobilenet_v2_0.35_224_tl.h5) | Float | 224x224x3 | 49.01 % |
202
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_tl/mobilenet_v2_0.35_224_tl_int8.tflite) | Int8 | 224x224x3 | 47.26 % |
203
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft.h5) | Float | 224x224x3 | 73.76 % |
204
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_0.35_224_fft/mobilenet_v2_0.35_224_fft_int8.tflite) | Int8 | 224x224x3 | 73.16 % |
205
+ | [MobileNet v2 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_1.0_224_fft/mobilenet_v2_1.0_224_fft.h5) | Float | 224x224x3 | 77.77 % |
206
+ | [MobileNet v2 1.0 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/food-101/mobilenet_v2_1.0_224_fft/mobilenet_v2_1.0_224_fft_int8.tflite) | Int8 | 224x224x3 | 77.09 % |
207
+
208
+
209
+ ### Accuracy with person dataset
210
+
211
+ The person dataset is derived from COCO-2014 and created using the script here (link). The dataset folder has 2 sub-folders — person and notperson containing images of the respective types
212
+ Dataset details: [link](https://cocodataset.org/) , License [Creative Commons Attribution 4.0](https://creativecommons.org/licenses/by/4.0/legalcode), Quotation[[3]](#3) , Number of classes: 2 , Number of images: 84810
213
+
214
+ | Model | Format | Resolution | Top 1 Accuracy |
215
+ |------------|--------|-----------|----------------|
216
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs.h5) | Float | 128x128x3 | 92.56 % |
217
+ | [MobileNet v2 0.35 tfs](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_tfs/mobilenet_v2_0.35_128_tfs_int8.tflite) | Int8 | 128x128x3 | 92.44 % |
218
+ | [MobileNet v2 0.35 tl ](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl.h5) | Float | 128x128x3 | 92.28 % |
219
+ | [MobileNet v2 0.35 tl](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_tl/mobilenet_v2_0.35_128_tl_int8.tflite) | Int8 | 128x128x3 | 91.63 % |
220
+ | [MobileNet v2 0.35 fft ](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft.h5) | Float | 128x128x3 | 95.37 % |
221
+ | [MobileNet v2 0.35 fft](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/ST_pretrainedmodel_public_dataset/person/mobilenet_v2_0.35_128_fft/mobilenet_v2_0.35_128_fft_int8.tflite) | Int8 | 128x128x3 | 94.95 % |
222
+
223
+
224
+ ### Accuracy with ImageNet
225
+
226
+ Dataset details: [link](https://www.image-net.org), License: BSD-3-Clause, Quotation[[4]](#4)
227
+ Number of classes: 1000.
228
+ To perform the quantization, we calibrated the activations with a random subset of the training set.
229
+ For the sake of simplicity, the accuracy reported here was estimated on the 50000 labelled images of the validation set.
230
+
231
+ | Model | Format | Resolution | Top 1 Accuracy |
232
+ |----------|--------|------------|----------------|
233
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_128/mobilenet_v2_0.35_128.h5) | Float | 128x128x3 | 46.96 % |
234
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_128/mobilenet_v2_0.35_128_int8.tflite) | Int8 | 128x128x3 | 43.94 % |
235
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_224/mobilenet_v2_0.35_224.h5) | Float | 224x224x3 | 56.44 % |
236
+ | [MobileNet v2 0.35](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_0.35_224/mobilenet_v2_0.35_224_int8.tflite) | Int8 | 224x224x3 | 54.7 % |
237
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224.h5) | Float | 224x224x3 | 68.87 % |
238
+ | [MobileNet v2 1.0](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8.tflite) | Int8 | 224x224x3 | 67.97 % |
239
+ | [MobileNet v2 1.0_per_tensor](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.0_224/mobilenet_v2_1.0_224_int8_per_tensor.tflite) | Int8 | 224x224x3 | 64.53 % |
240
+ | [MobileNet v2 1.4](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.4_224/mobilenet_v2_1.4_224.h5) | Float | 224x224x3 | 71.97 % |
241
+ | [MobileNet v2 1.4](https://github.com/STMicroelectronics/stm32ai-modelzoo/image_classification/mobilenetv2/Public_pretrainedmodel_public_dataset/ImageNet/mobilenet_v2_1.4_224/mobilenet_v2_1.4_224_int8.tflite) | Int8 | 224x224x3 | 71.46 % |
242
+
243
+
244
+ ## Retraining and Integration in a simple example:
245
+
246
+ Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
247
+
248
+
249
+ # References
250
+
251
+ <a id="1">[1]</a>
252
+ "Tf_flowers : tensorflow datasets," TensorFlow. [Online]. Available: https://www.tensorflow.org/datasets/catalog/tf_flowers.
253
+
254
+ <a id="2">[2]</a>
255
+ J, ARUN PANDIAN; GOPAL, GEETHARAMANI (2019), "Data for: Identification of Plant Leaf Diseases Using a 9-layer Deep Convolutional Neural Network", Mendeley Data, V1, doi: 10.17632/tywbtsjrjv.1
256
+
257
+ <a id="3">[3]</a>
258
+ L. Bossard, M. Guillaumin, and L. Van Gool, "Food-101 -- Mining Discriminative Components with Random Forests." European Conference on Computer Vision, 2014.
259
+
260
+ <a id="4">[4]</a>
261
+ Olga Russakovsky*, Jia Deng*, Hao Su, Jonathan Krause, Sanjeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg and Li Fei-Fei.
262
+ (* = equal contribution) ImageNet Large Scale Visual Recognition Challenge.